Streaming And Augmenting Stereo Camera Images - Visual Field Estimation using Oculus

Jayanth Ramesh and Pulkit Agrawal

Introduction

Problem Statement

Our objective for the project was to estimate the visual field of patients suffereing from various kinds of visual complications resulting from macular degeneration, retinitis pigmentosa, glaucoma, hypovolemia, hallucinogenic drugs, merucry poisoning, altitude sickness and other such causes in a efficient manner. We propose two algorithms for estimating the visual speed. One algorithm involves scanning the visual field first using a circular stimulus, followed by moving a point stimulus in a linear fashion. The second algorithm is a fast way to estimate the visual field, though the method is less accurate compared to the first algorithm. We call the first algorithm as Circular Linear Scan algorithm and the second algorithm as Mouse Tracking algorithm.

Motivation

Current Perimetry tests can take upto 23 minutes per eye for visual field estimation. This can become very tiring for the patient. Moreover, the tests are passive in the sense that they don't involve an active participation from the patient. In current tests, the visual field is scanned by flashing a light at different points in the field. The question was whether it is possible to reduce the time it takes to perform perimetry tests, so that it doesn't weigh heavily on the patients and can it be more interactive, in the sense that it involves an active participation from the patient?

Perimetry Tests

There are four major kind of perimetry tests.^[1]

Tangent Screen

The vision is tested by presenting different sized pins attached to a black wand which can be moved against a black background. The stimulus here (the pins) can be white or coloured.

Goldmann Perimeter

Goldmann perimeter is a popular method for performing perimetry tests. In this test, a hollow spherical white bowl is positioned at a set distance from the patient. The simulus, which is bright light, is moved from beyond the edge of the visual field into the visual field. Goldmann perimeter can test the entire range of peripheral vision. The brightness and intensity of the test light can be varied. The test is administered by an examiner. It has been used for a long time for detecting vision changes in Glaucoma patients which has been replaced by automated perimetry test in recent times.

Automated Perimetry

In automated perimetry, a mobile stimulus is moved by a perimetry machine. The machine flashes light at different points in the field and the patient indicates if the light is visible or not by pressing a button. The chin of the patient rests on the machine and the eye not being tested is covered. The patient is asked to fixate on a point at the center. Depending on the response of the patient, the computer automatically maps and calculates the patient's visual field. Automated Perimetry is the most commonly used form of perimetry for estimating visual field these days.

Microperimeter

It is a fine detail mapping technique that compensates for eye movement while fixating during the test. This is also a type of automated perimetry as the visual field is determined in a computerized fashion.

Stimulus Presentation

There are two popular methods of stimulus presentation, which can be categorized as static perimetry and kinetic. ^[2]

Static Perimetry

In this test, different locations of the field are tested by gradually increasing the brightness of light at each location until it becomes visible. This method is then repeated for several locations until entire visual field is mapped.

Kinetic Perimetry

Here, instead of presenting light of varying intensity at a given location, a stimulus of fixed intensity and size is moved across different locations in the field. This test is then repeated for lights of different intensity to get the overall visual field estimate.

Types of Visual Field Defect

Before moving on to describing our proposed algorithms for perimetry tests, we describe the different kinds of visual field defects that could be possible in a patient suffering from Macular Degeneration or Glaucoma. The types of defects can be grouped under one of the following five classes.^{[3] [4]
[5]
[6]
[7]
[8]}

Concentric Loss

This visual defect involves loss of vision in an annular region in the eye. Among the patients with concentric restriction of visual field, the most common disorders include nyctalopia (night blindness), caused by widespread loss of rod function and tapetoretinal degeneration. One of the problems with this type of visual defect is it is often noted only after the defect has developed into an advanced stage because numerous compensation strategies are used by the patients. The following images show possible forms of concentric losses.

Central Vision Loss

Tunnel Loss refers to loss of vision in the central area of the visual field. Macular Degeneration is one of the most common causes of this kind of visual defect. It generally affects elderly population but can affect young people in the form of Stargardt's disease, Best's disease, Inverse Retinitis Pigmentosa or such others.

Tunnel Vision

Tunnel vision is the constriction of visual field, resulting in the loss of peripheral vision with retention of central vision, sometimes creating a sensation of seeing through a narrow tube like a tunnel. Glaucoma, severe catracts, retinitis pigmentosa and eye strokes are some of the major causes leading to this kind of defect. Other causes can include brain damage, neurological damage such as optic neuritis, concussions and such others.

Sector Loss

We define sector loss to be lack of vision in a wedge-shaped area in the visual field, where the peak of the sector points towards the center of the visual field. Disturbances of choroidal blood supply are likely causes of this type of visual defect.

Off-centered Patch

We call off-centered patch to be degeneration of vision in a region of the visual field not involving the center of vision. This type of defect is possible because scotomas do not always occur in the center of vision. Glaucoma, Macular Degeneration, Diseases of optic brain, optic tracks can result in scotomas, which could be off-centered.

Proposed Algorithms

We propose two algorithms for performing the perimetry tests in an efficient manner. We name the first algorithm as Circular Linear Scan (alternatively called Keyboard Trace algorithm, since the responses of the patients are recorded through keyboard for tracing out the visual field), as it involves performing a scan using a circular stimulus followed by moving the stimulus in a linear fashion. The second algorithm is mouse-tracking algorithm (also called mouse trace algorithm as the mouse pointer is used to record inputs by the patient to trace out the estimate of the visual field), where the patient moves a mouse pointer across and indicates by a click when the mouse pointer becomes invisible

Circular Linear Scan

The algorithm has two major steps. The first step involves starting with the largest possible circular stimulus and decreasing its radius to the minimum possible value. At each stage, the patient indicates if a circle is fully visible or partially visible or completely invisible. The second step involves moving a point stimulus along a ray at different angles starting from the center. We record the locations at which the point becomes visible after being invisible or vice-versa. The response of the patient from the two steps are then combined to get an estimate the visual field.

Circular Scan

Start with the circle of largest possible radius
Keep reducing the radius of circle till the minimum possible radius and record the response of the patient at each step
The possible reponses of the patient include
- Fully visible - The entire circle is visible (we will denote this respnse by F)
- Partially visible - Only a part of the circle is visible (we will denote this response by P)
- Invisible - No portion of the circle is visble (this response is denoted by I)

Linear Scan

Scan the field linearly along different angles, starting from the center by moving a point stimulus
Record the locations where the patient indicated the point stimulus becoming invisible after being visible or coming into visibility after being invisible

Determining the type of Visual defect

Once we have recorded the user response, we first determine the type of visual defect, and then generate the estimate of the visual field. The type of visual defect can be determined from the responses to the circular scan part of the algorithm in the following manner.

Record the patient's response as the radius of the circle is reduced from the largest possible value to least possible
Now we have an array of responses, each element of the array corresponds to either 'F', 'P' or 'I'
If two consecutive elements in the sequence are same, merge them and replace them by one
Repeat the above step till no further reduction is possible
- For example, if the array is [F,F,F,P,P,I], the final reduced array is [F,P,I]
If the final reduced response sequence is (F,I,F), (F,P,I,F), (F,P,I,P,F), (F,I,P,F), then the patient is suffereing from concentric loss
If the final reduced response sequence is (F,I), (F,P,I) then the patient is suffereing from central vision loss
If the final reduced response sequence is (I,F), (I,P,F) then the patient is suffereing from tunnel vision
If the final reduced response sequence is (F,P), (P) then the patient is suffereing from sector vision loss
If the final reduced response sequence is (F,P,F), (P,F) then the patient is suffereing from off-center visual loss

Estimating the perimeter of Visual field

Determining the visual field estimate from linear scan varies slightly based on the defect type. The final visual field estimate is generated from the set of locations recorded based on the patient's reponse.

Linear Scan for Concentric Loss

From the circular scan, we know r1 and r2

Move the dot along different angles starting from the corresponding point on the circle with radius r2
For every ray, record the position where the dot becomes invisible after being visble and vice-versa
Stop scanning along a ray when the dot hits a circle of radius r1

Linear Scan for Central Vision Loss

From the circular scan, we know r1 and radius of the largest circle within which there is no vision at all (let's call its radius as r2)

Move the dot along different angles starting from the corresponding point on the circle with radius r2
For every ray, record the position where the dot becomes invisible after being visble and vice-versa
Stop scanning when the dot hits the circle of radius r1

Linear Scan for Tunnel Vision

From the circular scan, we know r1 and radius of the smallest circle beyond which all points are completely invisible (let's call its radius as r2)

Move the dot along different angles starting from the corresponding point on the circle with radius r1
For every ray, record the position where the dot becomes invisible after being visble and vice-versa
Stop scanning along a ray when the dot hits the circle of radius r2

Linear Scan for Sector Vision Loss

From the circular scan, we know the radius of the smallest circle beyond which all points are completely visible (let's call its radius as r1)

Move the dot along different angles starting from the center
For every ray, record the position where the dot becomes invisible after being visble and vice-versa
Stop scanning along a ray when the dot hits the circle of radius r1

Linear Scan for Off-centered Patch

From the circular scan, we know r1 and r2

Move the dot along different angles starting from the corresponding point on the circle with radius r1
For every ray, record the position where the dot becomes invisible after being visible and vice-versa
Stop scanning along a ray when the dot hits the circle of radius r2

Optimizating the Circular Linear Scan

We observed that we can make the linear scan part of the algorithm more efficient if we record additional information from the patient, which is to take note of the quadrants in which the circle is completely visible and in which it isn't, whenever the patient indicates parital visibility of the circular stimulus. We designate V as the set of quadrants where the circle is completely visible and N as the set of quadrants where the circle isn't completely visible. We apply the following optimization for different types of visual defects.